simarora/ConcurrentQA
收藏资源简介:
ConcurrentQA是一个文本多跳问答基准,要求跨多个数据分布(如维基百科和电子邮件数据)进行并发检索。该数据集由斯坦福大学和FAIR的研究人员构建,遵循了HotpotQA的数据收集过程和模式。该基准可用于研究检索中的泛化能力以及跨多个隐私范围(如公共维基百科文档和私人电子邮件)进行推理时的隐私问题。
ConcurrentQA is a textual multi-hop question answering benchmark that necessitates concurrent retrieval across multiple data distributions, such as Wikipedia and email datasets. This dataset was constructed by researchers from Stanford University and FAIR, following the data collection workflow and schema of HotpotQA. This benchmark can be utilized to study the generalization ability in retrieval tasks as well as the privacy issues that emerge when conducting reasoning across multiple privacy scopes, including public Wikipedia documents and private email materials.
数据集概述
名称: ConcurrentQA
类型: 文本多跳问答基准
目的: 该数据集旨在要求在多个数据分布(如Wikipedia和电子邮件数据)上进行并发检索,用于研究检索中的泛化能力以及跨多个隐私范围(公共Wikipedia文档和私人电子邮件)进行推理时的隐私问题。
构建者: 由斯坦福大学和FAIR的研究人员构建
数据收集和架构: 遵循HotpotQA的数据收集过程和架构
任务:
- 主要任务: 问答(Question-Answering)
- 相关任务: 检索(Retrieval),相关数据集可在此处找到:ConcurrentQA-Retrieval
语言: 英语(en)
许可证: MIT
数据下载命令: bash cd .. mkdir corpora cd corpora
wget https://dl.fbaipublicfiles.com/concurrentqa/corpora/enron_only_corpus.json wget https://dl.fbaipublicfiles.com/concurrentqa/corpora/combined_corpus.json wget https://dl.fbaipublicfiles.com/concurrentqa/corpora/wiki_only_corpus.json wget https://dl.fbaipublicfiles.com/concurrentqa/corpora/title2sent_map.json
相关资源:
- 模型训练和结果分析代码可在ConcurrentQA GitHub仓库找到。
引用: bibtex @article{arora2023reasoning, title={Reasoning over Public and Private Data in Retrieval-Based Systems}, author={Simran Arora and Patrick Lewis and Angela Fan and Jacob Kahn and Christopher Ré}, year={2023}, journal={Transactions of the Association for Computational Linguistics}, }




